Mouse embryonic fibroblasts exhibit extensivedevelopmental and phenotypic diversityPrabhat K. Singhala,b, Slim Sassia,c, Lan Lana,b, Patrick Aud, Stefan C. Halvorsena, Dai Fukumurad, Rakesh K. Jaind,1,and Brian Seeda,b,1

aCenter for Computational and Integrative Biology, Massachusetts General Hospital, Boston, MA 02114; bDepartment of Genetics, Harvard Medical School,Boston, MA 02115; cDepartment of Orthopedics, Harvard Medical School, Boston, MA 02115; and dSteele Laboratory of Tumor Biology, Department ofRadiation Oncology, Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114

Contributed by Rakesh K. Jain, November 16, 2015 (sent for review August 22, 2015; reviewed by Patricia A. D’Amore, Massimo Pinzani, and David Tuveson)

Analysis of embryonic fibroblasts from GFP reporter mice indicates thatthe fibroblast cell type harbors a large collection of developmentallyand phenotypically heterogeneous subtypes. Some of these cellsexhibit multipotency, whereas others do not. Multiparameter flowcytometry analysis shows that a large number of distinct populationsof fibroblast-like cells can be found in cultures initiated from differentembryonic organs, and cells sorted according to their surface pheno-type typically retain their characteristics on continued propagation inculture. Similarly, surface phenotypes of individual cloned fibroblast-like cells exhibit significant variation. The fibroblast cell class appears tocontain a very large number of denumerable subtypes.

embryo | fibroblast | lineage | heterogeneity | phenotype

Fibroblasts are ubiquitous mesenchymal cells that are thoughtto play important roles in wound repair and healing, and that

serve as reservoirs of multipotent progenitors capable of repopu-lating depleted cell compartments. The overgrowth of fibroblasts andelaboration of their secreted matrix products contributes to a varietyof human fibrotic diseases (1). In the setting of tumor biology, thecontributions of fibroblasts to tumor stroma have been recognized asimportant factors in the evolution of neoplastic tissue (2–6).Embryos are a rich source of fibroblasts, and murine embryonic

fibroblasts (MEFs) are frequently prepared to study the physio-logical consequences of selective gene ablations. MEFs are typicallyprepared by trypsinization and seeding of embryos into cell culturemedium after removal of the head, tail, limbs, and internal organs.The cultures that result after a few passages are often consideredlargely homogeneous populations.MEF cultures are widely believed to proliferate in response to

mitogenic factors in serum that arise from the platelet degranulationresulting from coagulation of whole blood. Among the growth fac-tors that have been identified from platelet releasates are TGF-β1,PDFG-β, angiopoietin 1, ECM protein 1, and vascular endothelialgrowth factor (VEGF)-C (7). Several of these growth factors arethought to play important roles in proliferation of tumor stromalfibroblasts (6).Previous studies have characterized a line of transgenic mice

expressing green fluorescent protein (GFP) under the control of afragment from the promoter region of VEGF-A. GFP-positive cellsseen to infiltrate tumors implanted in the transgenic mice wereidentified as fibroblasts, and MEFs isolated from transgenic mouseembryos were found to give rise to both GFP-positive and GFP-negative cells (8). MEF cultures consist of highly heterogeneouspopulations of distinct cell types. In the present study, we examinedthe origins of MEF heterogeneity in greater detail.

ResultsFibroblast-Like Populations from VEGF-GFP Transgenic Mice. Cells derivedfrom tumor stroma or from primary MEF cultures of VEGF-GFPtransgenic mice contain both GFP-positive and GFP-negative cellsresembling fibroblasts (8). Fibroblast-like cells isolated from VEGF-GFP mouse embryos of various ages ranging from embryo (post-coital) day E12.5 to E18.5 are morphologically indistinguishable by

light or electron microscopy, but exhibit variable GFP expression.Populations from earlier stages, days E12.5 and E14.5, exhibit lowerfrequencies of GFP-positive cells compared with cultures preparedfrom day E16.5 and E18.5 embryos (Fig. 1A). GFP-positive cellstypically replicate more rapidly than GFP-negative cells under stan-dard culture conditions, leading to an increase in their prevalencewith increasing culture passage (Fig. 1B).Following fluorescence-activated cell sorting of newly initiated

cultures, GFP-positive and GFP-negative cells can be propagated asapparently homogeneous populations. The GFP-positive cells re-semble cells from the immortalized fibroblast line NIH 3T3, whereasthe GFP-negative cells retain the characteristics of the cells initiallyplaced in culture and resemble the C3H10T1/2 cell line (SI Appendix,Fig. S1A). On electron microscopy (Fig. 1C), both cell types showtypical characteristics of fibroblasts: fairly smooth plasma membraneand nuclear membrane surfaces with delicate cytoplasmic processes,abundant intracytoplasmic ribosomes and undilated or dilated roughendoplasmic reticulum, and occasional mitochondria and pinocyticvesicles. No tight gap junctions are apparent between cells. Analysisof cytokine expression showed that the GFP-positive cells expressedVEGF-A, whereas the GFP-negative cells did not, consistent with thestructure of the transgene (SI Appendix, Fig. S1B). These data showthat fibroblast populations can be heterogeneous with respect toexpression in a way that is not immediately apparent, and raise thequestion of whether additional heterogeneity can be found.

Significance

Embryonic fibroblasts have long been considered a relatively ho-mogeneous cell type, and many studies have been conducted onfibroblasts of varying provenance. The results of the present studyindicate that embryonic fibroblasts are highly heterogeneous andexhibit anatomic and developmental variation that has both practi-cal consequences for the interpretation of experimental results andtheoretical implications for the nature and organization ofmetazoanorganisms. Studies using embryonic fibroblasts to explore the con-sequences of altered alleles should take into account the possibilitythat fibroblast composition may be altered by allele expression.

Author contributions: P.K.S., S.S., L.L., P.A., S.C.H., D.F., R.K.J., and B.S. designed research; P.K.S.,S.S., L.L., P.A., and S.C.H. performed research; P.K.S., S.S., L.L., P.A., S.C.H., D.F., R.K.J., and B.S.analyzed data; and P.K.S. and B.S. wrote the paper.

Reviewers: P.A.D., Schepens Eye Research Institute; M.P., University College London; andD.T., Cold Spring Harbor Laboratory.

The authors declare a conflict of interest. R.K.J. has received consultant fees from Oph-thotech, SPARC, SynDevRx, and XTuit; owns equity in Enlight, Ophthotech, SynDevRx, andXTuit; and serves on the Board of Directors of XTuit and the Board of Trustees of TeklaHealthcare Investors, Tekla Life Sciences Investors, Tekla Healthcare Opportunities Fund,and Tekla World Healthcare Fund. No reagents or funding from any these companieswere used in this study.

Freely available online through the PNAS open access option.1To whom correspondence may be addressed. Email: jain@steele.mgh.harvard.edu orbseed@ccib.mgh.harvard.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1522401112/-/DCSupplemental.

122–127 | PNAS | January 5, 2016 | vol. 113 | no. 1 www.pnas.org/cgi/doi/10.1073/pnas.1522401112

Analysis of Fibroblast-Like Cells by Immunofluorescence. In a flowcytometry survey, although the GFP-positive cells were relativelyhomogeneous, GFP-negative cells exhibited substantial heterogene-ity. Most of the GFP-negative population did not express the cellsurface proteins typical of committed hematopoietic, endothelial, orepithelial lineages (SI Appendix, Table S1). The pattern of expressionalso diverged from that reported for various compilations of mes-enchymal stem cell (MSC) surface proteins (9–14).GFP-negative cells variably expressed CD14, CD24, CD34, and

podoplanin (PDPN; Pdpn); expressed little CD73 and CD105; anddid not express CD133. In contrast, MSCs have not been reported toexpress CD14, CD24, CD34 or PDPN, but have been shown to ex-press CD73, CD105, and CD133 (SI Appendix, Table S1). Sca-1(Ly6a) has been found to be expressed by bone marrow-derivedMSCs (15), and the cells in this study were also positive for thismarker (SI Appendix, Table S1). Other markers of pluripotency, in-cluding Oct-4 (Pou5f1), Sox-2 (Sox2), Nanog (Nanog), Myc (Myc),nucleostemin (NST; Gnl3), nestin (Nes), Cox-2 (Cox2), Pax-7 (Pax7),Fra-1 (Fosl1), and Klf-4 (Klf4) were confirmed to be expressed bymurine ES cells (SI Appendix, Fig. S2A), whereas NIH 3T3 fibroblastsdid not express most of these markers (SI Appendix, Fig. S2B). Ex-pression of NST and Fra-1 was observed in NIH 3T3 cells, inagreement with earlier findings (16, 17). GFP-negative cells werefound to express nucleolar NST and nestin, along with limitedquantities of nuclear Fra-1, but none of the other proteins charac-teristic of pluripotent cells (SI Appendix, Fig. S3A). GFP-positive cellsdid not differ greatly from the GFP-negative cells in this respect (SIAppendix, Fig. S3B). Nestin and NST expression was higher in GFP-negative cells, whereas Fra-1 expression was higher in GFP-positive

cells. In addition, the GFP-negative cells did not express the em-bryonic stem cell markers CD31, SSEA1, SSEA3, and podocalyxin(Tra-1-81) (SI Appendix, Table S1).Some classification of fibroblast subtypes has been achieved by

examining the pattern of expression of contractile and intermediatefilament proteins, such as alpha-smooth muscle actin (αSMA; Acta2),smooth muscle protein 22-α (sM22α; Tagln), desmin (Des), vimentin(Vim), and myosin heavy chain (smMHC; Myh11) in fibroblasts,myofibroblasts, and smooth muscle cells (18). Both GFP-positive andGFP-negative cells were positive for αSMA, and sM22α (smoothmuscle or myofibroblast markers) and for vimentin and negative forsmMHC, a marker of more mature muscle cells, and desmin, asmooth muscle cell protein not normally found in fibroblasts ormyofibroblasts (SI Appendix, Fig. S4 A and B).Two candidate reagents that have been used to identify fibroblasts

are anti-FAPα, (Fap), which recognizes a member of the dipeptidylpeptidase family expressed by tumor stromal fibroblasts (19), andER-TR7, an antibody raised against thymic stromal cells that rec-ognizes an unidentified intracellular antigen (20, 21). GFP-positivecells express the ER-TR7 antigen and FAPα (SI Appendix, Fig.S4A), whereas GFP-negative cells express only FAPα (SI Appendix,Fig. S4B).The results of the foregoing studies show that GFP-negative

fibroblasts can be further divided into several subpopulations.The patterns of expression appear to be diverse and distinctive,suggesting either that multiple cell types are present or that anotherwise homogeneous population can display wide variation asa possible consequence of other influences, such as growthconditions or duration of culture.

Multiparameter Analysis Exposes a Highly Heterogeneous Population.Because the heterogeneity observed for individual cell surface pro-teins in the GFP-negative population could reflect the diversity of asingle population, each member of which could show variation, orcould arise from the presence of multiple subpopulations with distinctpatterns of expression, we conducted a multiparameter cytometryexperiment. A culture that was homogeneous by morphology wasdivided by expression of CD73 and CD146 (two proteins that reliablydemarcate subsets), and three resulting subpopulations were analyzedfor CD54 and CD71 expression. Analysis of the latter showed at leastfour distinct expression patterns, corresponding to CD54 high andlow and CD71 high and low (Fig. 2A, 1–3). Analysis of CD24, CD80,and CD90.1 within these subgroups revealed additional subpopula-tions (Fig. 2A, 4–6, 7–9, and 10–12).To identify whether the observed variation represented the mul-

tiple manifestations of a single cell type capable of facile conversionbetween phenotypes or the presence of a complex population of cellsstably expressing distinct patterns of cell surface proteins, we sub-jected selected subpopulations to fluorescence-activated cell sortingand initiated cultures from the sorted cells. We then reanalyzed thecultures for their sort markers after four to six passages. Most of theexpanded subpopulations retained their original surface phenotype,and in settings in which drift toward a less distinct expression patternwas observed, the pattern of drift was typically specific to the pop-ulation observed (Fig. 2B, 1–5). After CD54−, CD73−, and Sca-1+

cells were sorted for CD24, CD80, and CD90.1 expression andpropagated in culture, ∼60% of the CD24−, CD80−, CD90.1− cellsand ∼86% of the CD24+, CD80−, CD90.1+ cells retained their initialpattern of expression (Fig. 2B, 1 and 3). These data suggest that theobserved heterogeneity of fibroblast-like cells largely reflects theexpression programs of phenotypically stable cell types.

Single Colonies Give Rise to Different Phenotypes. If the cells of theMEF population stably express distinct cell surface phenotypes,then it should be possible to isolate clonally derived progeny thatexhibit those phenotypes. To test this, we plated MEFs from dayE18.5 embryos in medium containing 20% calf serum to yieldvisible colonies of various sizes at 6–8 wk in culture. Isolated

A

C

B

Fig. 1. Phenotypic comparisons of GFP-negative andGFP-positive cells. Fibroblastsfrom the VEGF-GFP mouse embryos at E12.5, E14.5, E16.5, and E18.5 were isolatedand cultured as described in Materials and Methods. (A) Flow cytometry analysesof GFP-positive and -negative cells. (B) Fibroblasts from E12.5 embryos at passage28 (1) or from E18.5 embryos at passage 30 (2) were analyzed cytometrically or byepifluorescence microscopy (3 and 4). (C) Representative electron micrographs ofGFP-negative cells (1 and 2; E12.5 from VEGF-GFP embryos) and GFP-positive cells(3 and 4; E18.5 embryos) at 2,000× magnification. Both GFP-positive and GFP-negative cell types exhibited characteristics of fibroblast-like cells.

Singhal et al. PNAS | January 5, 2016 | vol. 113 | no. 1 | 123

CELL

BIOLO

GY

colonies were individually picked and cultured in 96-wellmicroplates. The serum percentage was gradually reduced to 10%as the clones expanded. Failures of expansion were common, andmany of the colonies did not fill wells of a 96-well plate or sub-sequently did not fill wells of a 48-well plate, with some showingsigns of senescence. Out of 227 colonies with normal morphology,20 clones could be expanded (SI Appendix, Table S2). Each of theclones was analyzed by microscopy and cytometry for GFP status(SI Appendix, Fig. S5 A and B). Three GFP-positive cell populationswere obtained, a proportion roughly consistent with the observedpercentage of GFP-positive cells in day E18.5 cultures. Individualclones could acquire characteristics of fat, muscle, and bone fol-lowing exposure to the appropriate media (SI Appendix, Fig. S6).Consistent with the interpretation that the starting population

was heterogeneous, surface phenotyping of cells derived fromGFP-negative colonies revealed great variability (Fig. 2C and SIAppendix, Table S3). These data suggest that the phenotypicdiversity observed in unfractionated populations could be retainedby single cells and their progeny through many rounds of cel-lular division.

Single-Cell Transcriptome Analysis. Single-cell transcriptome datawere collected from 136 unfractionated cells of an E12.5 embryo, 86fibroblasts from E18.5 embryo skeletal muscle, and 88 cells from asorted and cultured MEF population (CD80−, Sca-1−, CD24−,CD38−, CD90−, CD107a−). Analysis of principal components yielded10 statistically significant components that could be reduced to twodimensions using t-distributed stochastic neighbor embedding (tSNE)(22–24). Density clustering with differential expression analysis gaveseven transcriptionally distinct clusters ranging in size from 5 to 96cells (Fig. 3A). Populations 2–7 are classified as distinct, whereaspopulation 1 is composed of cells dissimilar to one another and thecells of populations 2–7. Populations 2 and 4 are the largest, andalthough they are classified as single populations computationally,elements of diversity that do not rise to the level of statistical sig-nificance are clearly present (Fig. 3B). Of the markers used for flowcytometry analysis, only five (CD80, CD38, CD24, PDPN, and Sca-1)were found to be expressed at levels adequate for analysis, and foreach the expression levels were widely distributed, consistent with theobserved cytometric heterogeneity (Fig. 3C). The genes shown in theheat map are listed in SI Appendix, Table S4 and a more compre-hensive list of the highly variable genes is presented in SI Appendix,Table S5. These experiments show that profound differences in

transcriptional profile can be detected among single fibroblast cells,even when a single organ type serves as the fibroblast source.

Organotypic Expression of Fibroblast Surface Phenotypes. To determinewhether embryonic location contributes to the diversity of fibroblastsubtypes, we isolated fibroblasts from multiple embryonic organs.Cells from each organ were subjected to multiparameter flowcytometry without being subjected to culture conditions. FAPα wasused to gate a population that could be considered distinct from cellsof neuroepithelial origin. Cells expressing FAPα were present inall organs, albeit to varying degrees (SI Appendix, Fig. S7). Subpop-ulations characterized by differential expression of CD38 andCD140a were further analyzed for CD24 and Sca-1 expression, whichrevealed a variety of distinguishable fibroblast subtypes (Fig. 4A and

A

CD54 CD80 CD24 CD24

1C

D90

CD

90

CD

90

3

2

4 5

10

6

12

CD

24

CD

38

CD

80

CD

105

CD

90

CD

106

Sca

1P

DP

N

A2 B2 B6 C1 C4

B3

CD24

CD

90

1

4

2

5

A1C

7 8 9C

D90

CD

90

CD

90

CD

80 C

D8 0

C

D80

CD54 CD80 CD24 CD24

CD54 CD80 CD24 CD24

CD

71 C

D71

C

D71

11

Fig. 2. Heterogeneity within fibroblast-like cells.(A) Subset analyses. First column (1–3), CD71 and CD54subsets; subsequent columns, analysis of the CD71+

CD54− subset by CD24, CD80, and CD90. Top row (1, 4,5, 6), CD146+ CD73+ population; middle row (2, 7, 8,9), CD146− CD73+ population; bottom row (3, 10, 11,12), CD146+ CD73− population. (B) Stability of sur-face phenotype. Cultures initiated from subpopu-lations were reanalyzed after propagation in culturefor four to six passages. Top row (1–3), reanalysis ofCD54− CD73− Sca+ CD80− cells that were originallyCD24− CD90− (1), CD24− CD90+ (2), or CD24+ CD90+

(3). Bottom row (4 and 5), CD54− CD73− Sca+ CD80+

cells, originally CD24− CD90− (4) or CD24− CD90+ (5).(C) Heterogeneity of cell populations derived fromisolated colonies. Individual colonies were picked andexpanded, and the surface expression patterns of theprogeny were analyzed by flow cytometry.

A

C

B

Fig. 3. Single-cell analyses identify transcriptionally distinct clusters. (A) A tSNEplot of seven distinct populations. Cell populations range from the very distinctand small (e.g., population 5) to the larger populations with potential furthersubdivisions (e.g., population 2). (B) Heat map of gene expression by markersthat best differentiate the seven cell populations. (C) Violin density plots in theseven cell populations for CD80, CD38, CD24, PDPN, and Sca-1. The ordinate isthe gene expression level measured in counts per million mapped reads. Eachdot represents an expression value for one cell. The width is a smoothedprobability density of the expression values.

124 | www.pnas.org/cgi/doi/10.1073/pnas.1522401112 Singhal et al.

SI Appendix, Fig. S9). Organ-specific patterns of embryonic fibroblastsubsets were readily detected.We also initiated fibroblast cultures from 13 embryonic organs.

Most cultures were almost entirely FAPα-positive (SI Appendix,Fig. S8), showing that FAPα can be expressed by fibroblasts in situor in culture. A significant proportion of the fibroblast-like cellscultured from heart did not express FAPα, however. The culturedfibroblasts from various organs also displayed significant diversity(SI Appendix, Fig. S9).

GFP-Negative Cells Display Multipotency, But GFP-Positive Cells Do Not.GFP-negative cells isolated from day E12.5 and E18.5 embryoswere exposed to culture conditions that promote in vitro differen-tiation. Adipogenic differentiation was marked by an accumulationof intracellular fat droplets visualized by oil red O staining after aninduction period of 2 wk (SI Appendix, Fig. S10A, 1). QuantitativePCR (qPCR) of cDNA prepared from differentiated culturesdemonstrated an induction of adipocyte-specific genes (SI Appen-dix, Fig. S10 B, 6 and C, 6). No change in transcripts encodingcaldesmon (taken to be a protein exhibiting invariant expression)was observed.Osteogenic differentiation was evaluated by cytochemical staining

with alizarin red for calcium deposition or alcian blue to visualizeproteoglycan deposition, and for alkaline phosphatase (SI Appendix,Fig. S10A, 2–4). Osteogenic transcription patterns were revealedby qPCR evaluating the expression of osteoblast-specific markers(SI Appendix, Fig. S10 B, 5 and C, 5).Myogenic differentiation was evaluated morphologically, by

syncytium formation observed after 2–3 wk of culture. Confocalimmunofluorescence microscopy showed greatly increased smoothmuscle myosin heavy chain (Myh11) immunoreactivity comparedwith cells propagated under conventional conditions (SI Appendix,Figs. S4B and S10A, 9). qPCR demonstrated increased expressionof muscle cell markers. (SI Appendix, Fig. S10 B, 4 and C, 4).GFP-positive cells did not differentiate into adipogenic or osteo-

genic cells, as assessed by cytochemical staining (SI Appendix, Fig.S10A). In addition, qPCR did not reveal any significant shiftsin transcript abundance characteristic of the respective lineages(SI Appendix, Fig. S10 B, 1–3 and C, 1–3). Expression of CD44,CD80, CD120a, and CD121 was reduced after induction of differ-entiation in all cases, whereas the densities of CD24, CD34, andCD106 were reduced in some, but not all, conditions (SI Appendix,Fig. S11). Expression of CD29, CD90.1, CD140a, Sca-1, and PDPNwere little affected, and a CD36 population appeared in cultures ofcells exposed to adipogenic conditions.

GFP-Negative Cells Can Form Pericyte-Like Structures in Vivo andContract Collagen in Vitro. Pericytes, cells that girdle the abluminalside of endothelial cells in the perivascular space, may be the mostconsequential cell type in the microvascular complications of di-abetes (25, 26). The potential of the cell population to differentiateinto pericytes has been explored in a mouse cranial window modelthat permits study of the microvasculature and associated cell typesin vivo by intravital microscopy (27). Here, unlabeled GFP-negativecells and GFP-labeled human umbilical vein endothelial cells(HUVECs) were coimplanted in mouse cranial windows (28). Atvarious intervals, the windows were examined by intravital mi-croscopy to observe the ability of HUVECs to form complexmicrovessels. Newly formed microvessels were observed withgreen HUVECs (Fig. 4B, 1 and 2). Next, GFP-negative cells werelabeled with dsRED (SI Appendix, Fig. S12A) and coimplantedwith unlabeled HUVECs. Red fluorescent cells were observedaround newly formed vessels (Fig. 4B, 3 and 4), suggesting thedifferentiation of the implanted cells into pericyte-like structures.To further assess the multifunctional capacity of the GFP-negative

cells in vitro, we performed collagen gel contraction assays. Whencultured in a collagen type I gel in serum-free media for 1–2 d, GFP-negative cells reduced the diameter of the gel to<50% of the original

size (SI Appendix, Fig. S12B). NIH 3T3 fibroblasts and GFP-positivecells did not produce significant contraction of the collagen gel.Using a matrigel culture support and serum-free medium, GFP-negative cells (SI Appendix, Fig. S12C, 1) and 10T1/2 cells adopted asmooth muscle morphology after only 1–3 d in culture (SI Appendix,Fig. S12C, 2). In comparison, GFP-positive cells at passage 6 or 34did not show smooth muscle morphology (SI Appendix, Fig. S12C, 3and 4). After detachment from the matrigel and immunofluores-cence analysis for smooth muscle-specific caldesmon, nearly all cellswere immunoreactive (SI Appendix, Fig. S12D, 1–4). These resultsdemonstrate that fibroblast-like cells from embryos have the ca-pacity to mature into cell types similar to those of pericytes.

Intestine Liver Bladder Kidney Stomach

140a

-38+

140

a+38

- 14

0a-3

8-

140a

+38+

Tail A

Brain Muscle skin Lung Heart Spleen Spinal cord

140a

-38+

140

a+38

- 14

0a-3

8-

B 1

2 4

3

Sca

-1

CD-24

Fig. 4. Diversity of FAPα-positive uncultured fibroblasts and multipotency ofGFP-negative cells. (A) Heterogeneity among fibroblast-like cells from embryonicorgans. Isolated cells were not cultured before analysis. Fibroblasts from severalorgans obtained from E18.5 embryos were analyzed using antibodies recogniz-ing FAPα, CD140a, CD38, Sca-1, and CD24. FAPα-positive cells were selected forflow cytometry studies. (B) Two-photon micrographs of mouse cranial windows.(1 and 2) Windows were coimplanted with unlabeled GFP-negative fibroblastsand GFP-labeled HUVECs. Newly formed vessels were visualized with i.v.-injectedtetramethylrhodamine dextran and HUVECs encircling them are in green, ondays 39 (1) and 109 (2). (3 and 4) Windows were coimplanted with dsRED- la-beled GFP-negative fibroblasts and unlabeled HUVECs at day 60, leading todsRED-labeled cells surrounding the newly formed vessels.

Singhal et al. PNAS | January 5, 2016 | vol. 113 | no. 1 | 125

CELL

BIOLO

GY

DiscussionThis study establishes that embryonic cells of the fibroblast classconstitute a rich population of subtypes. From a transgenic reportermouse line in which a VEGF promoter fragment drives the expressionof GFP, cultures containing subpopulations of GFP-positive and GFP-negative cells with the typical characteristics of fibroblasts could beinitiated. Within the GFP-negative cells, further subdivisions could bemade based on the abundance of various cell surface proteins. Ouranalysis identified a large number of distinct subpopulations in culturesinitiated from embryos. Cells with differing cell surface phenotypeswere not easily distinguished from one another by morphology; forexample, despite obvious differences in functionality, GFP-negativeand GFP-positive fibroblasts were not distinguishable ultrastructurally.GFP-negative fibroblasts displayed a well-spread morphology

with prominent stress fibers and could be induced to differentiateinto cell types of fat, muscle, and bone lineages. These features arealso exhibited by the fibroblast-like cells identified as MSCs (29);however, the patterns of phenotypic expression identified here di-verge from previously identified compilations of MSC phenotypes.MSCs from bone marrow have been purified by negative selectionfor antibodies against CD34, CD45, and CD14. In contrast, wefound that GFP-negative cells expressed detectable levels of CD34and CD14. GFP-negative cells show low levels of CD105 and CD73,in contrast to MSCs. MSCs have generally been shown to expressSca-1, PDGR-α (CD140a), NST, and CD133 (9, 13, 15, 30), whereasGFP-negative cells do not express CD133 and express variable levelsof PDGR-α. GFP-negative cells express CD24, unlike preparationsof MSCs. GFP-negative cells express PDPN, a marker suggested tobe absent from MSCs (10, 11). The pattern of surface proteinmarkers of the GFP-negative cells does not coincide with that ofother known MSCs or the more primitive multipotent adult pro-genitor cells (14).Although GFP-negative cells express some proteins found in ES

cells, such as NST, nestin, and Fra-1, they do not express other stemcell markers, such as Sox-2, Klf-4, Oct-4, cMyc, CD31, SSEA-1,SSEA-3, and Tra-1-81. Both GFP-positive and GFP-negative cellsexpress proteins found in contractile and intermediate filaments, suchas αSMA, sM22α, and vimentin, but lack others, such as desmin andsmMHC. GFP-positive cells express ER-TR7 and FAPα, whereasGFP-negative cells express only FAPα.Following the identification of a small number of surface markers

that could be used to subdivide the population by expression pattern,the fibroblast-like cells were found to be extremely heterogeneous,belying what appear to be relatively widespread assumptions aboutMEF uniformity. Multicolor flow cytometry revealed a complexpopulation structure of subtypes with varying degrees of stability ofthe cell surface phenotype. Gating cells by CD73 and CD146 ex-pression gave three subtypes that were further analyzed for CD54and CD71 expression, producing 12 distinct patterns. Additionalgating on CD24, CD80, and CD90.1 allowed further differentiationof subtypes. On expansion of sorted populations in culture, the mostfrequently observed behavior was retention of the expression patternused to select the population.Consistent with the observed retention of characteristics following

culture of sorted subpopulations, cells from individual coloniesshowed unimodal abundance distributions and distinguishablecharacteristics, a finding inconsistent with the interpretation thatthe heterogeneous expression may be attributed to gene or genenetwork noise (31). Cells with differing surface phenotypes couldbe coerced into bone, muscle, and fat lineages, indicating that atleast some of the cells exhibited true multipotency.Single-cell expression profiling revealed intraorgan variation in

cells captured from striated muscle. Multiparameter flow cytometricanalysis of FAPα-positive cells from various organs showed that thediversity of fibroblast expression is exhibited by cells in vivo.Moreover, distinct patterns of expression were associated with dif-ferent organs, indicating at least some form of organotypic gene

expression. The use of FAPα as a marker was validated in part bythe finding that a majority of fibroblasts growing in culture fromvarious organ sources expressed FAPα.Heterogeneity is an attribute of multiple cell types, including many

sources of MSCs (32–36), hematopoietic progenitors (37, 38), adultcardiac side populations (39), and side populations of skeletalmuscle (40). The extent to which this heterogeneity is transitory,representing gene or gene network noise, and to what extent it isstable, representing epigenetic subtypes, is not clear. For example,embryonic stem cell cultures show heterogeneity of expression of thetranscription factors Nanog (41), Gata6, Rex1 (42), and Dppa3 (43),but the retention of pluripotency and the typically high fraction ofclonal ES cell lines that can fully reconstitute all embryonic lineagessupport the view that the heterogeneity represents gene or genenetwork noise rather than variegated lineage-specific expression.Fibroblast reprogramming has applications in drug discovery, dis-

ease modeling, tissue engineering, and regenerative medicine. Theidentification of extensive heterogeneity within fibroblast populationssuggests that the developmental trajectories and possibly final statesof the reprogrammed cells may vary considerably. Because reprog-ramming attempts often succeed for only a very small fraction of thecells targeted for fate reassignment, the possibility that the reprog-rammed cells arise from a distinct subpopulation with unusual at-tributes should be considered.From a practical perspective, the study of the phenotypic

consequences of genetic changes is not easily carried out usingfibroblasts unless specific steps are taken to guard against effectsarising from inadvertent comparison of two cell lines of differentfibroblast subtypes. One practice that can protect against this isthe use of two sublines of a cloned fibroblast line from a nullmouse, one subline consisting of the original (null) clone of fi-broblast cells transduced with an empty retroviral vector and theother consisting of the null cell clone transduced with a retroviralvector that restores the missing gene product (44). If precautionsof this type are not taken, there is a possibility that the variationobserved between embryonic fibroblasts isolated from differentmice could reflect the properties of different populations of fi-broblasts, imperiling the conclusions reached.Because the principal method for assigning subtypes in this

study relied on cytometric classification of cell surface proteinexpression patterns, there is a reasonable likelihood that addi-tional patterns will become apparent when less-biased methodsare applied. At a minimum, the subclasses defined by single-celltranscriptome analysis do not obviously correspond to the sub-types defined by surface marker expression, and the identifica-tion of distinct patterns of expression in different tissues suggeststhat additional organ-specific characters may be present. Al-though a precise estimate of the number of fibroblast typespresent in vivo cannot be made, the possibility of tens to hun-dreds of subtypes seems plausible.

Materials and MethodsPreparation of MEF Cell Cultures. All animal studies were conducted followingprotocols approved by the Institutional Animal Care and Usage Committee ofMassachusetts General Hospital. Embryos from VEGF-GFP transgenic micewereisolated between E12.5 and E18.5. After the heads, tails, limbs, andmost of theinternal organs were removed, the embryos were minced and typsinized for20 min, and then seeded into T-75 cell culture dishes in 15 mL of complete MEFmedia (SI Appendix, Materials andMethods). The cells were split at 1:2–1:3 ratioswhen freshly confluent, passaged two or three times to obtain a morphologi-cally homogenous culture, and then frozen or expanded for further studies.

Immunofluorescence. For multicolor flow cytometry of fibroblasts from em-bryonic organs, extracted organs were minced into smaller pieces anddigested using Liberase DL (Roche) and DNase I (Roche) in HBSS (SI Appendix,Materials and Methods) in a CO2 incubator at 37 °C for 20–30 min. Digestedtissues were passed through a 100-μmmesh to obtain single-cell suspensions.Fc receptors were blocked, and cells were stained for flow cytometry anal-yses. Antibodies are listed in SI Appendix, Table S7.

126 | www.pnas.org/cgi/doi/10.1073/pnas.1522401112 Singhal et al.

Single-Cell Capture, mRNA Library Preparation, and Sequencing. Cells were sus-pended at 250–500 cells/μL. The cell suspension and Fluidigm C1 suspensionreagent were mixed at a 3:2 ratio. Then 20 μL of mix was loaded into a FluidigmC1 integrated fluidic circuit system. RNA extraction and mRNA amplificationwere performed for single cells within the C1 system using the ClontechSMARTer system. cDNA products were prepared for sequencing using an Illu-mina Nextera XT-DNA sample preparation kit. The R package Seurat (45) wasused to generate cell population clusters. Principal component analysis wasperformed on genes present in at least five cells with an expression value of onecount per million mapped reads. Individual cells were projected based on theirprincipal component scores onto a 2D plot using tSNE (24).

Mouse Cranial Window Preparation, Implantation, and Two-Photon Microscopy.Mouse cranial windows were prepared as described previously (46). Tetra-methylrhodamine dextran was used to visualize newly formed vessels.

More detailed information on themethods and reagents used in this studyis provided in SI Appendix, Materials and Methods.

ACKNOWLEDGMENTS. We thank Sylvie Roberge for assistance with thecranial window preparation, Peigen Huang for management of the mousecolony, and Gino Ferraro for comments on the manuscript. This researchwas supported by National Institutes of Health Grants P01CA080124,R35CA197743, and R01DK103879.

1. Schuppan D, Pinzani M (2012) Anti-fibrotic therapy: Lost in translation? J Hepatol 56(Suppl 1):S66–S74.

2. Dvorak HF, Weaver VM, Tlsty TD, Bergers G (2011) Tumor microenvironment andprogression. J Surg Oncol 103(6):468–474.

3. Ishii G, Ochiai A, Neri S (2015) Phenotypic and functional heterogeneity of cancer-associated fibroblast within the tumor microenvironment. Adv Drug Deliv RevS0169-409X(15)00160-X.

4. Öhlund D, Elyada E, Tuveson D (2014) Fibroblast heterogeneity in the cancer wound.J Exp Med 211(8):1503–1523.

5. Orimo A, Weinberg RA (2007) Heterogeneity of stromal fibroblasts in tumors. CancerBiol Ther 6(4):618–619.

6. Pickup M, Novitskiy S, Moses HL (2013) The roles of TGFβ in the tumour microenvi-ronment. Nat Rev Cancer 13(11):788–799.

7. Wijten P, et al. (2013) High precision platelet releasate definition by quantita-tive reversed protein profiling: Brief report. Arterioscler Thromb Vasc Biol 33(7):1635–1638.

8. Fukumura D, et al. (1998) Tumor induction of VEGF promoter activity in stromal cells.Cell 94(6):715–725.

9. Baddoo M, et al. (2003) Characterization of mesenchymal stem cells isolated frommurine bone marrow by negative selection. J Cell Biochem 89(6):1235–1249.

10. Conrad C, et al. (2009) Multipotent mesenchymal stem cells acquire a lymphendo-thelial phenotype and enhance lymphatic regeneration in vivo. Circulation 119(2):281–289.

11. Imaizumi Y, Amano I, Tsuruga E, Kojima H, Sawa Y (2010) Immunohistochemicalexamination for the distribution of podoplanin-expressing cells in developing mousemolar tooth germs. Acta Histochem Cytochem 43(5):115–121.

12. Lv FJ, Tuan RS, Cheung KM, Leung VY (2014) Concise review: The surface markers andidentity of human mesenchymal stem cells. Stem Cells 32(6):1408–1419.

13. Tondreau T, et al. (2005) Mesenchymal stem cells derived from CD133-positive cells inmobilized peripheral blood and cord blood: Proliferation, Oct4 expression, andplasticity. Stem Cells 23(8):1105–1112.

14. Reyes M, et al. (2002) Origin of endothelial progenitors in human postnatal bonemarrow. J Clin Invest 109(3):337–346.

15. Houlihan DD, et al. (2012) Isolation of mouse mesenchymal stem cells on the basis ofexpression of Sca-1 and PDGFR-α. Nat Protoc 7(12):2103–2111.

16. Gruda MC, Kovary K, Metz R, Bravo R (1994) Regulation of Fra-1 and Fra-2 phos-phorylation differs during the cell cycle of fibroblasts and phosphorylation in vitro byMAP kinase affects DNA binding activity. Oncogene 9(9):2537–2547.

17. Tsai RY, McKay RD (2002) A nucleolar mechanism controlling cell proliferation in stemcells and cancer cells. Genes Dev 16(23):2991–3003.

18. Abderrahim-Ferkoune A, et al. (2004) Transdifferentiation of preadipose cells intosmooth muscle-like cells: Role of aortic carboxypeptidase-like protein. Exp Cell Res293(2):219–228.

19. Scanlan MJ, et al. (1994) Molecular cloning of fibroblast activation protein alpha, amember of the serine protease family selectively expressed in stromal fibroblasts ofepithelial cancers. Proc Natl Acad Sci USA 91(12):5657–5661.

20. Van Vliet E, Melis M, Van Ewijk W (1984) Monoclonal antibodies to stromal cell typesof the mouse thymus. Eur J Immunol 14(6):524–529.

21. Van Vliet E, Melis M, Foidart JM, Van Ewijk W (1986) Reticular fibroblasts in peripherallymphoid organs identified by a monoclonal antibody. J Histochem Cytochem 34(7):883–890.

22. Amir AD, et al. (2013) viSNE enables visualization of high dimensional single-cell dataand reveals phenotypic heterogeneity of leukemia. Nat Biotechnol 31(6):545–552.

23. Macosko EZ, et al. (2015) Highly parallel genome-wide expression profiling of indi-vidual cells using nanoliter droplets. Cell 161(5):1202–1214.

24. van der Maaten L, Hinton G (2008) Visualizing data using t-SNE. J Mach Learn Res 9:2579–2605.

25. Arboleda-Velasquez JF, Valdez CN, Marko CK, D’Amore PA (2015) From pathobiologyto the targeting of pericytes for the treatment of diabetic retinopathy. Curr Diab Rep15(2):573.

26. Patel-Hett S, D’Amore PA (2011) Signal transduction in vasculogenesis and de-velopmental angiogenesis. Int J Dev Biol 55(4-5):353–363.

27. Yuan F, et al. (1994) Vascular permeability and microcirculation of gliomas andmammary carcinomas transplanted in rat and mouse cranial windows. Cancer Res54(17):4564–4568.

28. Koike N, et al. (2004) Tissue engineering: Creation of long-lasting blood vessels.Nature 428(6979):138–139.

29. Pittenger MF, et al. (1999) Multilineage potential of adult human mesenchymal stemcells. Science 284(5411):143–147.

30. Kafienah W, Mistry S, Williams C, Hollander AP (2006) Nucleostemin is a marker ofproliferating stromal stem cells in adult human bone marrow. Stem Cells 24(4):1113–1120.

31. Kaern M, Elston TC, Blake WJ, Collins JJ (2005) Stochasticity in gene expression: Fromtheories to phenotypes. Nat Rev Genet 6(6):451–464.

32. Ratajczak MZ, Kucia M, Majka M, Reca R, Ratajczak J (2004) Heterogeneous pop-ulations of bone marrow stem cells: Are we spotting on the same cells from thedifferent angles? Folia Histochem Cytobiol 42(3):139–146.

33. Okamoto T, et al. (2002) Clonal heterogeneity in differentiation potential of im-mortalized human mesenchymal stem cells. Biochem Biophys Res Commun 295(2):354–361.

34. Panepucci RA, et al. (2004) Comparison of gene expression of umbilical cord vein andbone marrow-derived mesenchymal stem cells. Stem Cells 22(7):1263–1278.

35. Vogel W, et al. (2003) Heterogeneity among human bone marrow-derived mesen-chymal stem cells and neural progenitor cells. Haematologica 88(2):126–133.

36. Colter DC, Sekiya I, Prockop DJ (2001) Identification of a subpopulation of rapidly self-renewing and multipotential adult stem cells in colonies of human marrow stromalcells. Proc Natl Acad Sci USA 98(14):7841–7845.

37. Chang HH, Hemberg M, Barahona M, Ingber DE, Huang S (2008) Transcriptome-widenoise controls lineage choice in mammalian progenitor cells. Nature 453(7194):544–547.

38. Morita Y, Ema H, Nakauchi H (2010) Heterogeneity and hierarchy within the mostprimitive hematopoietic stem cell compartment. J Exp Med 207(6):1173–1182.

39. Yamahara K, et al. (2008) Heterogeneic nature of adult cardiac side population cells.Biochem Biophys Res Commun 371(4):615–620.

40. Uezumi A, et al. (2006) Functional heterogeneity of side population cells in skeletalmuscle. Biochem Biophys Res Commun 341(3):864–873.

41. Singh AM, Hamazaki T, Hankowski KE, Terada N (2007) A heterogeneous expressionpattern for Nanog in embryonic stem cells. Stem Cells 25(10):2534–2542.

42. Toyooka Y, Shimosato D, Murakami K, Takahashi K, Niwa H (2008) Identification andcharacterization of subpopulations in undifferentiated ES cell culture. Development135(5):909–918.

43. Carter MG, et al. (2008) An in situ hybridization-based screen for heterogeneouslyexpressed genes in mouse ES cells. Gene Expr Patterns 8(3):181–198.

44. Zhou GL, Na SY, Niedra R, Seed B (2014) Deficits in receptor-mediated endocytosis andrecycling in cells from mice with Gpr107 locus disruption. J Cell Sci 127(Pt 18):3916–3927.

45. Satija R, Farrell JA, Gennert D, Schier AF, Regev A (2015) Spatial reconstruction ofsingle-cell gene expression data. Nat Biotechnol 33(5):495–502.

46. Brown E, Munn LL, Fukumura D, Jain RK (2010) In vivo imaging of tumors. Cold SpringHarb Protoc, 10.1101/pdb.prot5452.

Singhal et al. PNAS | January 5, 2016 | vol. 113 | no. 1 | 127

CELL

BIOLO

GY